[Qemu-devel] Re: Add target memory mapping API

[Mike Day]

On 19/01/09 19:28 +0200, Avi Kivity wrote:
> Ian Jackson wrote:
>>> Correct.  If you need to perform read-modify-write, you need to use  
>>> cpu_physical_memory_rw(), twice.  If we ever want to support RMW, 
>>> we'll need to add another value for is_write.  I don't think we have  
>>> interesting devices at this point which require efficient RMW.
>>>     
>>
>> Efficient read-modify-write may be very hard for some setups to
>> achieve.  It can't be done with the bounce buffer implementation.
>> I think ond good rule of thumb would be to make sure that the interface
>> as specified can be implemented in terms of cpu_physical_memory_rw.
>>   
>
> What is the motivation for efficient rmw?

rmw would be good for changing bytes in a device config space, or
programming an IOMMU. Right now these are not on the critical path but
they may be soon in some cases. For example imagine an assigned I/O
device that wants to dynamically map io translations for each DMA.


Mike

-- 
Mike Day
http://www.ncultra.org
AIM: ncmikeday |  Yahoo IM: ultra.runner
PGP key: http://www.ncultra.org/ncmike/pubkey.asc